WO2018209193A1 - Probability map-based ultrasound scanning - Google Patents
Probability map-based ultrasound scanning Download PDFInfo
- Publication number
- WO2018209193A1 WO2018209193A1 PCT/US2018/032247 US2018032247W WO2018209193A1 WO 2018209193 A1 WO2018209193 A1 WO 2018209193A1 US 2018032247 W US2018032247 W US 2018032247W WO 2018209193 A1 WO2018209193 A1 WO 2018209193A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- information
- target
- interest
- probability
- unit
- Prior art date
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/60—Analysis of geometric attributes
- G06T7/62—Analysis of geometric attributes of area, perimeter, diameter or volume
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
- A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B8/00—Diagnosis using ultrasonic, sonic or infrasonic waves
- A61B8/08—Detecting organic movements or changes, e.g. tumours, cysts, swellings
- A61B8/0833—Detecting organic movements or changes, e.g. tumours, cysts, swellings involving detecting or locating foreign bodies or organic structures
- A61B8/085—Detecting organic movements or changes, e.g. tumours, cysts, swellings involving detecting or locating foreign bodies or organic structures for locating body or organic structures, e.g. tumours, calculi, blood vessels, nodules
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B8/00—Diagnosis using ultrasonic, sonic or infrasonic waves
- A61B8/08—Detecting organic movements or changes, e.g. tumours, cysts, swellings
- A61B8/0866—Detecting organic movements or changes, e.g. tumours, cysts, swellings involving foetal diagnosis; pre-natal or peri-natal diagnosis of the baby
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B8/00—Diagnosis using ultrasonic, sonic or infrasonic waves
- A61B8/08—Detecting organic movements or changes, e.g. tumours, cysts, swellings
- A61B8/0883—Detecting organic movements or changes, e.g. tumours, cysts, swellings for diagnosis of the heart
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B8/00—Diagnosis using ultrasonic, sonic or infrasonic waves
- A61B8/08—Detecting organic movements or changes, e.g. tumours, cysts, swellings
- A61B8/0891—Detecting organic movements or changes, e.g. tumours, cysts, swellings for diagnosis of blood vessels
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B8/00—Diagnosis using ultrasonic, sonic or infrasonic waves
- A61B8/13—Tomography
- A61B8/14—Echo-tomography
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B8/00—Diagnosis using ultrasonic, sonic or infrasonic waves
- A61B8/42—Details of probe positioning or probe attachment to the patient
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B8/00—Diagnosis using ultrasonic, sonic or infrasonic waves
- A61B8/46—Ultrasonic, sonic or infrasonic diagnostic devices with special arrangements for interfacing with the operator or the patient
- A61B8/461—Displaying means of special interest
- A61B8/463—Displaying means of special interest characterised by displaying multiple images or images and diagnostic data on one display
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B8/00—Diagnosis using ultrasonic, sonic or infrasonic waves
- A61B8/52—Devices using data or image processing specially adapted for diagnosis using ultrasonic, sonic or infrasonic waves
- A61B8/5215—Devices using data or image processing specially adapted for diagnosis using ultrasonic, sonic or infrasonic waves involving processing of medical diagnostic data
- A61B8/5223—Devices using data or image processing specially adapted for diagnosis using ultrasonic, sonic or infrasonic waves involving processing of medical diagnostic data for extracting a diagnostic or physiological parameter from medical diagnostic data
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B8/00—Diagnosis using ultrasonic, sonic or infrasonic waves
- A61B8/52—Devices using data or image processing specially adapted for diagnosis using ultrasonic, sonic or infrasonic waves
- A61B8/5292—Devices using data or image processing specially adapted for diagnosis using ultrasonic, sonic or infrasonic waves using additional data, e.g. patient information, image labeling, acquisition parameters
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B8/00—Diagnosis using ultrasonic, sonic or infrasonic waves
- A61B8/56—Details of data transmission or power supply
- A61B8/565—Details of data transmission or power supply involving data transmission via a network
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N7/00—Computing arrangements based on specific mathematical models
- G06N7/01—Probabilistic graphical models, e.g. probabilistic networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
- G06T7/0014—Biomedical image inspection using an image reference approach
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H30/00—ICT specially adapted for the handling or processing of medical images
- G16H30/20—ICT specially adapted for the handling or processing of medical images for handling medical images, e.g. DICOM, HL7 or PACS
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H30/00—ICT specially adapted for the handling or processing of medical images
- G16H30/40—ICT specially adapted for the handling or processing of medical images for processing medical images, e.g. editing
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/43—Detecting, measuring or recording for evaluating the reproductive systems
- A61B5/4306—Detecting, measuring or recording for evaluating the reproductive systems for evaluating the female reproductive systems, e.g. gynaecological evaluations
- A61B5/4318—Evaluation of the lower reproductive system
- A61B5/4325—Evaluation of the lower reproductive system of the uterine cavities, e.g. uterus, fallopian tubes, ovaries
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/43—Detecting, measuring or recording for evaluating the reproductive systems
- A61B5/4375—Detecting, measuring or recording for evaluating the reproductive systems for evaluating the male reproductive system
- A61B5/4381—Prostate evaluation or disorder diagnosis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10132—Ultrasound image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30084—Kidney; Renal
Definitions
- Ultrasound scanners are typically used to identify a target organ or other structures in the body and/or determine features associated with the target organ/structure, such as the size of the organ/structure or the volume of fluid in the organ.
- ultrasound scanners are used to identify a patient's bladder and estimate the volume of fluid in the bladder.
- the ultrasound scanner is placed on the patient and triggered to generate ultrasound signals which comprise sound waves output at a specific frequency.
- the echoes from the ultrasound signals may be received by the scanner and analyzed to determine the volume of fluid in the bladder.
- the received echoes may be used to generate corresponding images that can be analyzed to detect boundaries of the target organ, such as the bladder wall.
- the volume of the bladder may then be estimated based on the detected boundary information.
- typical ultrasound scanners often suffer from inaccuracies caused by a number of factors, such as the variability of the size and/or shape of the target organ of interest from patient to patient, obstructions in the body that make it difficult to accurately detect boundaries of the target organ/structure, etc.
- Fig. 1 A illustrates an exemplary configuration of a scanning system consistent with an exemplary implementation
- Fig. IB illustrates operation of the scanning system of Fig. 1A with respect to detecting an organ in a patient
- Fig. 2 illustrates an exemplary configuration of logic elements included in the scanning system of Fig. 1A;
- Fig. 3 illustrates a portion of the data acquisition unit of Fig. 2 in an exemplary implementation
- Fig. 4 illustrates a portion of autoencoder unit of Fig. 2 in an exemplary
- Fig. 5 illustrates an exemplary configuration of components included in one or more of the elements of Fig. 2;
- Fig. 6 is a flow diagram illustrating processing by various components illustrated in Fig. 2 in accordance with an exemplary implementation
- Fig. 7 illustrates output generated by the autoencoder of Fig. 2 in an exemplary implementation
- Fig. 8 illustrates a binarization process in accordance with the processing of Fig. 6
- Fig. 9 is a flow diagram associated with displaying information via the base unit of Fig. 1A.
- Fig.10 illustrates exemplary image data output by the base unit in accordance with the processing of Fig. 9.
- Implementations described herein relate to using machine learning, including using neural networks and deep learning, to identify an organ or structure of interest in a patient based on information obtain via an ultrasound scanner.
- the scanner may be used to transmit a number of ultrasound signals toward the target organ and echo information associated with transmitted signals may be processed using machine learning
- the machine learning processing may be used to identify the target of interest and generate probability information associated with each portion or pixel of an image generated based on the received ultrasound echo data.
- ultrasound echo data such as B-mode echo data associated with ultrasound signals transmitted on a number of different scan planes directed to the target organ
- ultrasound echo data may be used to generate a probability map for each B-mode image.
- each pixel in the B-mode image may be mapped to a probability indicating whether that particular pixel is within or part of the target organ/structure.
- the result of the pixel-by -pixel analysis is used to generate a target probability map.
- a binarization process and post-processing may then be performed to remove noise and provide a more accurate representation of the organ, as compared to conventional scanners that attempt to determine boundary walls for the target organ and estimate the size based on the boundary information.
- the output from the post-processing is displayed to medical personnel and may aid in easily locating the organ while performing the ultrasound scan. Additional post-processing may also be performed to estimate a volume for the target organ, such as the volume of fluid in a patient's bladder.
- Fig. 1A is a diagram illustrating a scanning system 100 consistent with an exemplary embodiment.
- scanning system 100 includes probe 1 10, base unit 120 and cable 130.
- Probe 1 10 includes handle portion 112 (also referred to as handle 1 12), trigger 114 and nose portion 116 (also referred to as dome or dome portion 1 16). Medical personnel may hold probe 110 via handle 112 and press trigger 1 14 to activate one or more ultrasound transceivers and transducers located in nose portion 1 16 to transmit ultrasound signals toward the target organ of interest.
- Fig. IB illustrates probe 110 located on the pelvic area of patient 150 and over a target organ of interest, which in this example is the patient's bladder 152.
- Handle 1 12 allows a user to move probe 1 10 relative to patient 150.
- trigger 1 14 initiates an ultrasound scan of a selected anatomical portion while dome 116 is in contact with a surface portion of patient 150 when the selected anatomical portion is scanned.
- Dome 1 16 is typically formed of a material that provides an appropriate acoustical impedance match to the anatomical portion and/or permits ultrasound energy to be properly focused as it is projected into the anatomical portion.
- an acoustic gel or gel pads illustrated at area 154 in Fig. IB, may be applied to patient 150's skin over the region of interest (ROI) to provide an acoustical impedance match when dome 116 is placed against patient 150's skin.
- ROI region of interest
- Dome 116 includes one or more ultrasound transceiver elements and one or more transducer elements (not shown in Fig. 1A or IB).
- the transceiver elements transmit ultrasound energy outwardly from the dome 1 16, and receive acoustic reflections or echoes generated by internal structures/tissue within the anatomical portion.
- the one or more ultrasound transducer elements may include a one-dimensional, or a two-dimensional array of piezoelectric elements that may be moved within dome 1 16 by a motor to provide different scan directions with respect the transmissions of ultrasound signals by the transceiver elements.
- the transducer elements may be stationary with respect to probe 110 so that the selected anatomical region may be scanned by selectively energizing the elements in the array.
- probe 110 may include a directional indicator panel (not shown in Fig. 1 A) that includes a number of arrows that may be illuminated for initial targeting and guiding a user to access the targeting of an organ or structure within the ROI.
- a directional indicator panel (not shown in Fig. 1 A) that includes a number of arrows that may be illuminated for initial targeting and guiding a user to access the targeting of an organ or structure within the ROI.
- the directional indicator panel may be presented on display 122 of base unit 120.
- the one or more transceivers located in probe 110 may include an inertial reference unit that includes an accelerometer and/or gyroscope positioned preferably within or adjacent to dome 1 16.
- the accelerometer may be operable to sense an acceleration of the transceiver, preferably relative to a coordinate system, while the gyroscope may be operable to sense an angular velocity of the transceiver relative to the same or another coordinate system.
- the gyroscope may be of a conventional configuration that employs dynamic elements, or it may be an optoelectronic device, such as an optical ring gyroscope.
- the accelerometer and the gyroscope may include a commonly packaged and/or solid-state device.
- the accelerometer and/or the gyroscope may include commonly packaged micro-electromechanical system (MEMS) devices.
- MEMS micro-electromechanical system
- the accelerometer and gyroscope cooperatively permit the determination of positional and/or angular changes relative to a known position that is proximate to an anatomical region of interest in the patient.
- Probe 1 10 may communicate with base unit 120 via a wired connection, such as via cable 130. In other implementations, probe 1 10 may communicate with base unit 120 via a wireless connection (e.g., Bluetooth, WiFi, etc.).
- base unit 120 includes display 122 to allow a user to view processed results from an ultrasound scan, and/or to allow operational interaction with respect to the user during operation of probe 1 10.
- display 122 may include an output display/screen, such as a liquid crystal display (LCD), light emitted diode (LED) based display, or other type of display that provides text and/or image data to a user.
- display 122 may provide instructions for positioning probe 110 relative to the selected anatomical portion of patient 150. Display 122 may also display two-dimensional or three-dimensional images of the selected anatomical region.
- display 122 may include a graphical user interface (GUI) that allows the user to select various features associated with an ultrasound scan.
- GUI graphical user interface
- display 122 may allow a user to select whether patient 150 is male, female or a child.
- This allows system 100 to automatically adapt the transmission, reception and processing of ultrasound signals to the anatomy of a selected patient, such as adapt system 100 to accommodate various anatomical details of male and female patients.
- system 100 may be configured to locate a single cavity, such as a urinary bladder in the male patient.
- system 100 may be configured to image an anatomical portion having multiple cavities, such as a bodily region that includes a bladder and a uterus.
- system 100 may be configured to adjust the transmission based on the smaller size of the child patient.
- system 100 may include a cavity selector configured to select a single cavity scanning mode, or a multiple cavity-scanning mode that may be used with male and/or female patients. The cavity selector may thus permit a single cavity region to be imaged, or a multiple cavity region, such as a region that includes an aorta and a heart to be imaged.
- the selection of the type of patient e.g., male, female, child
- dome 1 16 may be positioned against a surface portion of patient 150 as illustrated in Fig. IB that is proximate to the anatomical portion to be scanned.
- the user actuates the transceiver by depressing trigger 114.
- the transducer elements optionally position the transceiver, which transmits ultrasound signals into the body, and receives corresponding return echo signals that may be at least partially processed by the transceiver to generate an ultrasound image of the selected anatomical portion.
- system 100 transmits ultrasound signals in a range that extends from approximately about two megahertz (MHz) to approximately 10 or more MHz (e.g., 18 MHz).
- probe 110 may be coupled to a base unit 120 that is configured to generate ultrasound energy at a predetermined frequency and/or pulse repetition rate and to transfer the ultrasound energy to the transceiver.
- Base unit 120 also includes one or more processors or processing logic configured to process reflected ultrasound energy that is received by the transceiver to produce an image of the scanned anatomical region.
- probe 110 may be a self-contained device that includes a microprocessor positioned within the probe 1 10 and software associated with the microprocessor to operably control the transceiver, and to process the reflected ultrasound energy to generate the ultrasound image.
- a display on probe 1 10 may be used to display the generated image and/or to view other information associated with the operation of the transceiver.
- the information may include alphanumeric data that indicates a preferred position of the transceiver prior to performing a series of scans.
- the transceiver may be coupled to a general-purpose computer, such as a laptop or a desktop computer that includes software that at least partially controls the operation of the transceiver, and also includes software to process information transferred from the transceiver so that an image of the scanned anatomical region may be generated.
- a general-purpose computer such as a laptop or a desktop computer that includes software that at least partially controls the operation of the transceiver, and also includes software to process information transferred from the transceiver so that an image of the scanned anatomical region may be generated.
- Fig. 2 is a block diagram of functional logic components implemented in system 100 in accordance with an exemplary implementation.
- system 100 includes data acquisition unit 210, convolutional neural network (CNN) autencoder unit 220, post processing unit 230, aiming logic 240 and volume estimating logic 250.
- probe 1 10 may include data acquisition unit 210 and the other functional units (e.g., CNN autoencoder unit 220, post processing unit 230, aiming logic 240 and volume estimating logic 250) may be implemented in base unit 120.
- CNN convolutional neural network
- probe 110 may transmit echo data and/or image data to a processing system via, for example, a wireless connection (e.g., WiFi or some other wireless protocol/technology) that is located remotely from probe 1 10 and base unit 120.
- a wireless connection e.g., WiFi or some other wireless protocol/technology
- probe 110 may include a transceiver that produces ultrasound signals, receives echoes from the transmitted signals and generates B-mode image data based on the received echoes (e.g., the magnitude or intensity of the received echoes).
- data acquisition unit 210 obtains data associated with multiple scan planes corresponding to the region of interest in patient 150.
- probe 1 10 may receive echo data that is processed by data acquisition unit 210 to generate two- dimensional (2D) B-mode image data to determine bladder size and/or volume.
- probe 110 may receive echo data that is processed to generate three- dimensional (3D) image data that can be used to determine bladder size and/or volume.
- FIG. 3 illustrates an exemplary data acquisition unit 210 used to obtain 3D image data.
- data acquisition unit 210 includes transducer 310, outer surface 320 of dome portion 1 16 and base 360. The elements illustrated in Fig. 3 may be included within dome portion 1 16 of probe 1 10.
- Transducer 310 may transmit ultrasound signals from probe 1 10, indicated by 330 in Fig. 3.
- Transducer 310 may be mounted to allow transducer 310 to rotate about two perpendicular axes.
- transducer 310 may rotate around a first axis 340 with respect to base 360 and rotate around a second axis 350 with respect to base 360.
- the first axis 340 is referred to herein as the theta axis
- the second axis 350 is referred to herein as the phi axis.
- the range of theta and phi motion may be less than 180 degrees.
- the scanning may be interlaced with respect to the theta motion and phi motion.
- movement of transducer 310 may occur in the theta direction followed by movement in the phi direction. This enables data acquisition unit 210 to obtain smooth continuous volume scanning as well as improving the rate at which the scan data is obtained.
- data acquisition unit 210 may resize the B-mode images prior to forwarding the image to CNN autoencoder unit 220.
- data acquisition unit 210 may include logic to reduce the size of the B-mode images through a reduction or decimation process. The reduced size B-mode images may then be input to CNN autoencoder unit 220, which will generate an output probability mapping, as described in more detail below.
- CNN autoencoder unit 220 may reduce or decimate the input B-mode image itself at the input layer. In either case, reducing the size/amount of B-mode image data may reduce the processing time and processing capability needed by CNN autoencoder unit 220 to process the B-mode image data.
- no resizing may be performed by data acquisition unit 210 prior to inputting the B-mode image data to CNN autoencoder unit 220.
- no resizing may be performed by data acquisition unit 210 prior to inputting the B-mode image data to CNN autoencoder unit 220.
- image enhancement operations such as brightness normalization, contrast enhancement, scan conversion may be performed by data acquisition unit 210 and/or CNN autoencoder unit 220 to improve accuracy with respect to generating output data.
- CNN autoencoder unit 220 may include logic for processing data received via data acquisition unit 210.
- CNN autencoder unit 220 may perform deep neural network (DNN) processing that includes a number of convolutional layer processings and a number of kernels or filters for each layer, as described in more detail below.
- DNN deep neural network
- CNN autoencoder unit or “autoencoder unit” as used herein should be broadly construed to include a neural network and/or machine learning system/unit in which both the input and output have spatial information, in contrast to classifiers that outputs global labels without spatial information.
- CNN autoencoder unit 220 includes logic that maps received image input to output with a least possible amount of distortion.
- CNN processing may be similar to other types of neural network processing, but CNN processing uses the explicit assumption that the inputs are images, which allows the CNN processing to more easily encode various properties/limitations into the processing, thereby reducing the amount of parameters that must be processed or factored by CNN autoencoder unit 220.
- CNN processing uses the explicit assumption that the inputs are images, which allows the CNN processing to more easily encode various properties/limitations into the processing, thereby reducing the amount of parameters that must be processed or factored by CNN autoencoder unit 220.
- CNN autoencoder unit 220 performs convolutional processing to generate features maps associated with the input image. The feature maps may then be sampled a number of times to generate an output.
- the kernel size of the CNN used by CNN autoencoder unit 220 may have a size of 17x17 or smaller to provide adequate speed for generating an output.
- the 17x17 kernel size allows CNN autoencoder unit 220 to capture adequate information around a point of interest within B- mode image data.
- the number of convolutional layers may be eight or less with five or less kernels for each layer.
- kernel sizes e.g., 3x3, 7x7, 9x9, etc.
- larger kernel sizes e.g., greater than 17x17
- additional kernels per layer e.g., greater than five
- additional convolutional layers e.g., more than ten and up to hundreds
- the data dimension is reduced by adding a narrow bottleneck layer within the processing such that only the data of interest can pass through the narrow layer.
- This data dimension reduction is typically accomplished by adding "pooling" layers or using a large “stride” to reduce the size of the image processed by the neural network.
- pooling and/or large stride is minimally used or combined with other spatial resolution-preserving techniques, such as skip connection or dilated
- system 100 may include other types of autoencoder units or machine learning units.
- CNN autoencoder unit 220 may include a neural network structure in which the output layer has the same number of nodes as the input layer.
- other types of machine learning modules or units may be used in which the size of the input layers does not equal the size of the output layers.
- a machine learning module may generate a probability mapping output that is two times larger or smaller (in terms of the number of layers) than the input image.
- a machine learning unit included in system 100 may use various machine learning techniques and algorithms, such as decision trees, support vector machines, Bayesian networks, etc. In each case, system 100 uses machine learning algorithms to generate probability information with respect to the B-mode input data that may then be used to estimate the volume of the target organ of interest, as described in detail below.
- Fig. 4 schematically illustrates a portion of CNN autoencoder unit 220 consistent with an exemplary implementation.
- CNN autoencoder unit 220 may include spatial input 410, FFT input 420, lookup 422, feature maps 430, feature maps 440, lookup 442, kernels 450, bias 452, kernels 460 and bias 462.
- Input spatial 410 may represent 2D B- mode image data provided by data acquisition unit 210.
- CNN autoencoder 220 may perform a Fast Fourier Transform (FFT) to convert the image data into a frequency domain, apply filters or weights to the input FFT via kernels FFT 450.
- FFT Fast Fourier Transform
- the output of the convolution processing may be biased via biasing value 452 and an inverse Fast Fourier Transform (IFFT) function applied with the result sent to look up table 422 to generate spatial feature maps 430.
- CNN autoencoder unit 220 may apply a FFT to spatial feature maps 430 to generate FFT feature maps 440 and the process may repeat for the additional convolutions and kernels. For example, if CNN autoencoder unit 220 includes eight convolutional layers, the process may continue seven more times.
- the kernels applied to each succeeding feature map correspond to the number of kernels times the number of feature maps, as illustrated by the four kernels 460 in Fig. 4.
- Biases 452 and 462 may also be applied to improve performance of the CNN processing.
- CNN autoencoder unit 220 may perform convolutions in the frequency domain using FFTs.
- FFTs Fast Fourier transforms
- system 100 may use a hand-held unit and base station, such as probe 110 and base unit 120, to perform CNN processing.
- a spatial-domain approach may be used.
- a spatial-domain approach may use additional processing power in situations where system 100 is able to communicate with other processing devices, such as with processing devices connected to system 100 via a network (e.g., a wireless or wired network) and/or operating with system 100 via a client/server approach (e.g., system 100 is the client).
- the output of CNN autoencoder unit 220 is probability information associated with a probability that each processed portion or pixel of the processed input image is within the target organ of interest.
- CNN autoencoder unit 220 may generate a probability map in which each pixel associated with the processed input image data is mapped to a probability corresponding to a value between 0 and 1 , where the value zero represents 0% probability that the pixel is within the target organ and the value one represents 100% probability that the pixel is within the target organ, as described in more detail below.
- CNN autoencoder unit 220 performs the pixel analysis or spatial location analysis on the processed images, as opposed to the input images.
- the pixel-by-pixel analysis of the processed images may not correspond on a one-to-one basis with the input images.
- one processed pixel or spatial location analyzed by CNN autoencoder unit 220 to generate probability information may correspond to multiple pixels in the input image, or vice versa, based on resizing of the input images.
- the term "probability” as used herein should be construed to broadly include a likelihood that a pixel or portion of an image is within a target or organ of interest.
- the term “probability information” as used herein should also be broadly construed to include discrete values, such as binary values or other values.
- CNN autoencoder unit 220 may generate a probability map in which each pixel is mapped to various values that can be correlated to probability values or indicators, such as values ranging from -10 to 10, values corresponding to one of 256 gray scale values, etc.
- the values or units generated by CNN autoencoder unit 220 may be used to determine the probability that a pixel or portion of an image is within the target organ.
- a value of one may indicate a 0% probability that a pixel or portion of an image is within the target organ and a value of 256 may indicate a 100% probability that a pixel or image is within the target organ.
- CNN autoencoder unit 220 may generate discrete output values, such as binary values, that indicate whether a pixel or output area is within the target organ.
- CNN autoencoder unit 220 may include a binarization or classification process that generates a discrete value, such as a "1" when the pixel is within the target organ and a "0" when the pixel is not within the target organ.
- the generated values may not be binary, but may correlate to whether the pixel is within the target organ or outside the target organ.
- CNN autoencoder unit 220 may take various factors into consideration when analyzing the pixel-by-pixel data. For example, CNN autoencoder unit 220 may receive input from a user via the GUI displayed on display 122 of base unit 120
- CNN autoencoder unit 220 may include thee different CNNs trained with male, female and child data and CNN autoencoder unit 220 may use the appropriate CNN based on the selection.
- CNN autoencoder unit 220 may automatically identify patient demographics of the subject, such as the gender, age, age range, adult or child status, etc., using, for example, the B-mode image data associated with the subject. CNN autoencoder unit 220 may also automatically identify clinical conditions of the subject using, for example, the B-mode image data, such as the body mass index (BMI), the body size and/or weight, etc. CNN autoencoder unit 220 may also automatically identify device information for a scan performed by system 100, such as position information of probe 110, aiming quality of probe 110 with respect to the target of interest, etc.
- BMI body mass index
- another processing device may perform the automatic detection of patient demographics, clinical conditions and/or device information using, for example, another neural network or other processing logic, and the output of the automatic determination may be provided as an input to CNN autoencoder unit 220.
- patient demographic information, clinical conditions and/or device information, patient data, etc. may be manually entered via, for example, display 122 of base unit 120 or via input selections on probe 110.
- the information automatically identified by CNN autoencoder unit 220 or manually input to CNN autoencoder unit 220/system 100 may be used to select an appropriate CNN for the processing of the image data.
- CNN autoencoder unit 220 may be trained with other information.
- CNN autoencoder unit 220 may be trained with patient data associated with the subject, which may include information obtained using the patient's medical history data as well as information obtained via a physical examination of the patient prior to scanning a target of interest.
- patient data may include a patient's medical history information, such as patient surgery history, chronic disease history (e.g., bladder disease information), previous images of the target of interest (e.g., previous images of the subj ect's bladder), etc., as well as data obtained via a physical examination of the patient/subject, such as pregnancy status, presence of scar tissue, hydration issues, abnormality in the target region (e.g., a bloated or distended abdomen), etc.
- the patient data may be input to system 100 via display 122 of base unit 120.
- the information automatically generated by CNN autoencoder unit 220 and/or another processing device, and/or information entered manually to system 100 may be provided as inputs to the machine learning processing performed by system 100 to aid in increasing the accuracy of data associated with the target of interest generated by system 100.
- autoencoder unit 220 may receive input information regarding the type of organ (e.g., bladder, aorta, prostate, heart, kidney, uterus, a blood vessel, amniotic fluid, a fetus, etc.) via the GUI provided on display 122, the number of organs, etc., being imaged and use an appropriate CNN trained in accordance with the selected organ.
- organ e.g., bladder, aorta, prostate, heart, kidney, uterus, a blood vessel, amniotic fluid, a fetus, etc.
- Post processing unit 230 includes logic to receive the pixel-by-pixel probability information and applies a "smart" binarization probability algorithm. For example, post processing unit 230 may perform interpolation to more clearly define contour details, as described in detail below. In addition, post processing unit 230 may adjust the output of CNN autoencoder unit 220 based on the subject type. For example, if a "child" is selected via the GUI on display 122 prior to initiating an ultrasound scan using probe 1 10, post processing unit 230 may ignore output from CNN autoencoder unit 220 that corresponds to a location that is deeper than a certain depth because the depth of the bladder within a child is typically shallow, due to the small size of a typical child.
- post processing unit 230 may determine based on the organ type, whether to select a single dominant region or multiple regions of interest. For example, if the organ type being scanned is the bladder, post processing unit 230 may select a single dominant region because there is only one bladder in the body. However, if the target is the pubic bone, post processing unit 230 may select up to two regions of interest, corresponding to the two sides of the pubic bone.
- Aiming logic 240 includes logic to determine whether the target organ is properly centered with respect to probe 110 during the ultrasound scanning.
- aiming logic 240 may generate text or graphics to guide the user in adjusting the location of probe 110 to obtain a better scan of the target organ.
- aiming logic 240 may analyze data from probe 110 and determine that probe 110 needs to be moved to the left on patient 150.
- aiming logic 240 may output text and/or graphics (e.g., flashing arrows) to display 122 to direct the user to move probe 110 in the appropriate direction.
- Volume estimating logic 250 may include logic to estimate the volume of the target organ. For example, volume estimating logic 250 may estimate the volume based on the 2D images generated by post processing unit 230, as described in detail below. In scenarios where 3D images are provided, volume estimating logic 250 may simply determine the volume of the target organ using the 3D images. Volume estimating logic 250 may output the estimated volume via display 122 and/or a display on probe 110.
- system 100 may include more or fewer logic units/devices than illustrated in Fig. 2.
- system 100 may include multiple data acquisition units 210 and multiple processing units that process the received data.
- system 100 may include additional elements, such as communication interfaces (e.g., radio frequency transceivers) that transmit and receive information via external networks to aid in analyzing ultrasound signals to identify a target organ of interest.
- communication interfaces e.g., radio frequency transceivers
- CNN autoencoder unit 220 may convert input images into probability information, generate intermediate mapping outputs, as described below, and also convert the intermediate outputs into, for example, volume information, length information, area information, etc. That is, a single neural network processing device/unit may receive input image data and output processed image output data along with volume and/or size information. In this example, a separate post processing unit 230 and/or volume estimating logic 250 may not be needed.
- any intermediate mapping outputs may or may not be accessible or visible to an operator of system 100 (e.g., the intermediate mappings may be part of internal processing not directly accessible/visible to the user). That is, the neural network included in system 100 (e.g., CNN autoencoder unit 220) may convert received ultrasound echo information and/or images and output volume information or other size information for the target of interest, while requiring no additional input or little additional input by the user of system 100.
- the neural network included in system 100 e.g., CNN autoencoder unit 220
- CNN autoencoder unit 220 may convert received ultrasound echo information and/or images and output volume information or other size information for the target of interest, while requiring no additional input or little additional input by the user of system 100.
- Fig. 5 illustrates an exemplary configuration of a device 500.
- Device 500 may correspond to, for example, a component of CNN autoencoder unit 220, post processing unit 230, aiming logic 240 and volume estimating logic 250.
- device 500 may include bus 510, processor 520, memory 530, input device 540, output device 550 and communication interface 560.
- Bus 510 may include a path that permits communication among the elements of device 500.
- all or some of the components illustrated in Fig. 5 may be implemented and/or controlled by processor 520 executing software instructions stored in memory 530.
- Processor 520 may include one or more processors, microprocessors, or processing logic that may interpret and execute instructions.
- Memory 530 may include a random access memory (RAM) or another type of dynamic storage device that may store information and instructions for execution by processor 520.
- Memory 530 may also include a read only memory (ROM) device or another type of static storage device that may store static information and instructions for use by processor 520.
- Memory 530 may further include a solid state drive (SDD).
- SDD solid state drive
- Memory 530 may also include a magnetic and/or optical recording medium (e.g., a hard disk) and its corresponding drive.
- Input device 540 may include a mechanism that permits a user to input information to device 500, such as a keyboard, a keypad, a mouse, a pen, a microphone, a touch screen, voice recognition and/or biometric mechanisms, etc.
- Output device 550 may include a mechanism that outputs information to the user, including a display (e.g., a liquid crystal display (LCD)), a printer, a speaker, etc.
- a touch screen display may act as both an input device and an output device.
- Communication interface 560 may include one or more transceivers that device 500 uses to communicate with other devices via wired, wireless or optical mechanisms.
- communication interface 560 may include one or more radio frequency (RF) transmitters, receivers and/or transceivers and one or more antennas for transmitting and receiving RF data via a network.
- Communication interface 560 may also include a modem or an Ethernet interface to a LAN or other mechanisms for communicating with elements in a network.
- RF radio frequency
- device 500 may include more or fewer devices than illustrated in Fig. 5.
- device 500 performs operations in response to processor 520 executing sequences of instructions contained in a computer-readable medium, such as memory 530.
- a computer-readable medium may be defined as a physical or logical memory device.
- the software instructions may be read into memory 530 from another computer- readable medium (e.g., a hard disk drive (HDD), SSD, etc.), or from another device via communication interface 560.
- HDD hard disk drive
- SSD etc.
- hard-wired circuitry such as application specific integrated circuits (ASICs), field programmable gate arrays (FPGAs), etc.
- ASICs application specific integrated circuits
- FPGAs field programmable gate arrays
- implementations described herein are not limited to any specific combination of hardware circuitry and software.
- Fig. 6 is a flow diagram illustrating exemplary processing associated with identifying a target of interest, as well as identifying parameters (e.g., volume) associated with the target of interest. Processing may begin with a user operating probe 110 to scan a target organ of interest. In this example, assume that the target organ is a bladder. It should be understood that features described herein may be used to identify other organs or structures within the body.
- a user may press trigger 1 14 and the transceiver included in probe 1 10 transmits ultrasound signals and acquires B-mode data associated with echo signals received by probe 1 10 (block 610).
- data acquisition unit 210 may transmit ultrasound signals on 12 different planes through the bladder and generate 12 B-mode images corresponding to the 12 different planes.
- the data may correspond to 2D image data.
- data acquisition unit 210 may generate 3D image data.
- data acquisition unit 210 may perform interlaced scanning to generate 3D images. In each case, the number of transmitted ultrasound signals/scan planes may vary based on the particular implementation.
- data acquisition unit 210 may reduce the size of the B-mode images before forwarding the B-mode data to CNN autoencoder unit 220.
- data acquisition unit 210 may reduce the size of the B- mode images by 10% or more.
- CNN autoencoder unit 220 receives 2D B-mode data and processes the data to remove noise from the received data.
- CNN autoencoder unit 220 may receive B-mode image data 710, with a dark area or region 712 corresponding to the bladder.
- the B-mode image data includes areas that are irregular or may appear unclear or fuzzy to a user.
- region 712 in Fig. 7 includes lighter areas within the perimeter of the bladder, as well as boundaries that are not distinct. Such noisy areas may make it difficult to accurately estimate the volume of the bladder.
- CNN autoencoder unit 220 performs a de-noising of the acquired B-mode image 710 by generating a target probability map (block 620).
- CNN autoencoder 220 may utilize CNN techniques to generate probability information with respect to each pixel in the input image.
- Base unit 120 may then determine whether the full cone data (i.e., all of the scan plane data) has been acquired and processed (block 630). For example, base unit 120 may determine whether all 12 B-mode images corresponding to 12 different scans through the bladder have been processed. If all the B-mode image data has not been processed (block 630 - no), base unit 120 controls motion to the next scan plane position (block 640) and processing continues to block 610 to process the B-mode image associated with another scan plane.
- the full cone data i.e., all of the scan plane data
- base unit 120 may determine whether all 12 B-mode images corresponding to 12 different scans through the bladder have been processed. If all the B-mode image data has not been processed (block 630 - no), base unit 120 controls motion to the next scan plane position (block 640) and processing continues to block 610 to process the B-mode image associated with another scan plane.
- base unit 120 may revise the probability map using 3D information (block 650). For example, CNN
- CNN autoencoder unit 220 may use stored assumption information regarding the 3D shape and size of the bladder based on whether the patient is male, female, a child, etc., to modify some of the probability information generated by CNN autoencoder unit 220, thereby effectively modifying the size and/or shape of the bladder. That is, CNN autoencoder unit 220, as described above, may use a CNN trained based on demographic information of the patient, clinical conditions of the patient, device information associated with system 100 (e.g., probe 110), patient data (e.g., patient medical history information and patient examination data) of the patient, etc.
- CNN autoencoder unit 220 may use a CNN trained with male patient data if patient 150 is male, use a CNN trained with female patient data if patient 150 is female, use a CNN trained with child data if patient 150 is a child, use a CNN trained based on the patient's age range, use a CNN trained with the patient's medical history, etc.
- system 100 may display the P-mode image data (block 660), such as image 720 illustrated in Fig. 7.
- base unit 120 may use the probability map to segment the target region via a binarization process (block 670).
- post-processing unit 230 may receive the output of CNN autoencoder unit 220 and resize (e.g., via interpolation), smooth and/or de- noise (e.g., via filtering) the probability mapping.
- the probability map may be resized through interpolation to a larger size to obtain better resolution and/or to recover, at least partially, the spatial resolution of the original B-mode image data that may have been reduced in size.
- a 2D Lanczos interpolation may be performed to resize the image associated with the target probability map.
- base unit 120 may perform a classification or binarization process to convert the probability information from probability mapping unit to binarized output data.
- post processing unit 230 may convert the probability values to binary values. When multiple candidate probability values are identified for a particular pixel, post processing unit 230 may select the most prominent value. In this manner, post processing unit 230 may apply some "smartness" to select the most likely value when multiple candidates are identified.
- Fig. 8 schematically illustrates an exemplary smart binarization process.
- image 810 illustrates an output from a pixel classification or probability map corresponding to a 2D ultrasound image in which the probability information is converted to gray scale images having various intensities.
- image 810 includes a gray area labeled 812 and gray areas labeled 814 that represent possible locations for portions of the bladder.
- Post processing unit 230 identifies the peak point or point within image 810 having the greatest intensity, as illustrated by cross-hairs 822 illustrated in image 820. Postprocessing unit 230 may then fill the region around the peak point for regions whose intensity are greater than a threshold intensity, as illustrated by region 832 in image 830.
- Postprocessing unit 230 may then fill the background, as illustrated by region 842 in image 840. Post processing unit 230 then fills any holes or open regions within the image, as illustrated in area 852 in image 850.
- the holes in region 842 may correspond to noisy regions or regions associated with some obstruction in patient 150. In this manner, post processing unit 230 identifies the most probable location and size for the bladder. That is, area 852 is considered to be part of the bladder of patient 150.
- post processing unit 230 may use information other than a peak intensity value within image 810.
- post processing unit 230 may use a peak value of a processed probability, such as a peak of a smoothed probability map, use multiple peak values to identify multiple filled regions, etc.
- post processing unit 230 may select a "dominant" region based on area, peak probability or averaged probability in each region.
- post processing unit 230 may use one or multiple seed points manually input by an operator via, for example, display 122, use an algorithm that generates one or more seed points, perform another type of thresholding that does not use seed points, etc., to identify regions of the patient's bladder.
- base unit 120 may output an image, such as image 720 illustrated in Fig. 7.
- image 720 includes region 722 corresponding to the bladder.
- the edges of bladder 722 are much more distinct than the boundaries in image 712, providing a much more accurate representation of the bladder.
- base unit 120 may use brightness values for each pixel and local gradient values for adjacent pixels, as well as statistical methods, such as a hidden Markov model and neural network algorithms (e.g., CNN) to generate the probability value for each pixel in the B-mode image and de-noise the B-mode data.
- CNN hidden Markov model and neural network algorithms
- Base unit 120 may then convert the segmentation results to a target volume (block 670).
- post processing unit 230 may sum the volumes of all voxels in 3D space that correspond to each valid target pixel in the binarized maps. That is, volume estimating logic 250 may sum the voxels in the 12 segmented target images to estimate the volume of the bladder. For example, the contribution or volume of each voxel can be pre-calculated and stored in a lookup table within base unit 120. In this case, volume estimating logic 250 may use the sum of the voxels as an index to the lookup table to determine the estimated volume. Volume estimating logic 250 may also display the volume via display 122 of base unit 120.
- volume estimating logic 250 may display the estimated volume of the bladder at area 724 in Fig. 7 (i.e., 135 milliliters (mL) in this example), which is output to display 122 of base unit 120.
- volume estimating logic 250 may display the volume information via a display on probe 1 10.
- Post processing unit 230 may also display the segmentation results (block 690). That is, post processing unit 230 may display 12 segments of the bladder via display 122 of base unit 120.
- system 100 may not perform a binarization process on the probability mapping information.
- CNN auto encoder unit 220 and/or post processing unit 230 may apply a look-up table to the probability mapping information to identify likely portion of the target organ of interest and display the output via display 122.
- probability mapping unit 230 may display information as it is generated in real time.
- Fig. 9 illustrates exemplary processing associated with providing additional display information to a user.
- post processing unit 230 may display probability mode information (referred to herein as P- mode) via display 122 as it is generated in real time (Fig. 9, block 910).
- Post processing unit 230 may also segment the target (block 920) and display the segmentation results with the B- mode images (block 930).
- Fig. 10 illustrates three B-mode images 1010, 1012 and 1014 and corresponding P-mode images 1020, 1022 and 1024. In other implementations, all 12 B-mode images and 12 corresponding P-mode images may be displayed.
- post processing unit 230 may provide outlines for the boundaries of the bladder displayed in each of the P-mode images.
- each of P-mode images 1020, 1022 and 1024 may include outlines in, for example, a different color or brighter color than the interior portions of the bladder, as illustrated in Fig. 10.
- Implementations described herein use machine learning to identify an organ or structure of interest in a patient based on information obtain via an ultrasound scanner.
- the machine learning processing may receive image data and generate probability information for each particular portion of the image (e.g., pixel) to determine the probability that the particular portion is within the target organ.
- Post processing analysis may further refine the probability information using additional information, such as the gender or age of the patient, the particular target organ, etc.
- the volume of the target organ may also be provided to the user, along with real time probability mode images.
- a target of interest such as a patient's bladder
- CNN processing to estimate a volume of the target (e.g., bladder).
- other organs or structures may be identified, and sizes or other parameters associated with the organs/structures may be estimated.
- the processing described herein may be used to identify and display a prostate gland, a kidney, a uterus, ovaries, an aorta, a heart, a blood vessel, amniotic fluid, a fetus etc., as well as particular features associated with these targets, such as volume and/or size-related measurements.
- additional size-related measurements may be generated. For example, length, height, width, depth, diameter, area, etc., of an organ or region of interest may be calculated. As an example, for a scan of an aorta, measuring the diameter of the aorta may be important in trying to identify an anomaly, such as an aneurysm. For a prostate scan, measurement of the width and height of the prostate may be needed.
- measurements such as length, height, width, depth, diameter, area, etc. may be generated/estimated using the machine learning processing described above. That is, the machine learning described above may be used to identify boundary walls or other items of interest and estimate the particular size-related parameter of interest to medical personnel.
- C- mode image data which typically includes a representation of the target of interest (e.g., bladder) formed in a plane oriented perpendicular to B-mode images may be used in other implementations.
- radio frequency (RF) or quadrature signals e.g., IQ signals
- CNN autoencoder unit 220 may be used as input to CNN autoencoder unit 220 to generate a probability output mapping associated with the target.
- system 100 may generate one probability map for the target organ of interest (e.g., the bladder), another probability map for the pubic bone/pubic bone shadow, and another probability map for the prostate. In this manner, more accurate representations of the internal organs of patient 150 may be generated, which may result in more accurate volume estimation for the target organ (e.g., the bladder).
- target organ of interest e.g., the bladder
- the pubic bone/pubic bone shadow e.g., the pubic bone/pubic bone shadow
- another probability map for the prostate e.g., more accurate representations of the internal organs of patient 150 may be generated, which may result in more accurate volume estimation for the target organ (e.g., the bladder).
- features described herein relate to performing a pixel-by -pixel analysis of B-mode image data.
- an edge map may be used instead of a pixel-by -pixel mapping.
- the edges of the target may be detected using CNN algorithms.
- a polygon coordinate approach may be used to identify discrete portions of the bladder and then connect the points.
- a contour edge tracking algorithm may be used to connect the points of the target organ.
- various inputs such as information indicating whether the patient is male or female, a child, etc.
- Other inputs to the probability mapping and/or binarization may also be used.
- a body mass index (BMI) BMI
- age or age range may be input to base unit 120 and base unit 120 may automatically adjust the processing based on the particular BMI, age or age range.
- Still other inputs to the probability mapping and/or binarization process such as the depth of each pixel, plane orientation, etc., may be used to improve accuracy of the output images and/or volume estimate generated by system 100.
- training data associated with various types of patients, men, women and children may be used to aid in generating the P-mode data.
- thousands or more of training data images may be used to generate the CNN algorithms used to process the B-mode input data to identify the target or interest.
- thousands or more images may be input or stored in base unit 120 to aid in modifying the output of CNN autoencoder unit 220. This may be particularly helpful in scenarios where expected obstructions, such as a pubic bone for a bladder scan, adversely affect the images.
- base unit 120 may store information regarding how to account for and minimize effects of the obstruction. CNN autoencoder unit 220 and/or post processing unit 230 may then more accurately account for the obstruction.
- features described herein refer to using B-mode image data as an input to
- echo data associated with transmitted ultrasound signals may include harmonic information that can be used to detect a target organ, such as the bladder.
- higher order harmonic echo information e.g., second harmonic or higher
- probe 110 may transmit ultrasound signals at multiple frequencies and echo information associated with the multiple frequencies may be used as input to CNN autoencoder unit 220 or other machine learning modules to detect a target organ and estimate volume, size, etc., of the target organ.
- multiple B-mode images at the fundamental frequency and multiple B-mode images at the higher order harmonic frequency or frequencies may be used as inputs to CNN autoencoder unit 220.
- fundamental frequency and harmonic frequency information may be pre-processed and used as inputs to CNN autoencoder unit 220 to aid in generating the probability map.
- the ratio between harmonics and fundamental frequency powers may be used as an input to the CNN autoencoder unit 220 to enhance the accuracy of the probability mapping.
- the post processing described above may use a second machine learning (e.g., CNN) algorithm to de-noise the image data and/or perform outline/edge tracking for the images.
- a second machine learning e.g., CNN
- higher dimensional image date (e.g., 2.5D or 3D) data may be input to CNN autoencoder unit 220.
- CNN autoencoder unit 220 may use B-mode images associated with several scan planes, as well as neighboring scan planes to improve accuracy.
- CNN autoencoder unit 220 may generate 12 probability maps for each of 12 scan planes and post processing unit 230 may use all 12 probability maps to generate 3D images based on the 12 probability maps (e.g., via a 3D flood-filling algorithm).
- a classification and/or binarization process may then be performed on the 2.5D or 3D images to generate, for example, 3D output images.
- logic may include hardware, such as one or more processors, microprocessor, application specific integrated circuits, field programmable gate arrays or other processing logic, software, or a combination of hardware and software.
Landscapes
- Health & Medical Sciences (AREA)
- Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Public Health (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Molecular Biology (AREA)
- Radiology & Medical Imaging (AREA)
- Pathology (AREA)
- Animal Behavior & Ethology (AREA)
- Surgery (AREA)
- Veterinary Medicine (AREA)
- Heart & Thoracic Surgery (AREA)
- Theoretical Computer Science (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Epidemiology (AREA)
- Primary Health Care (AREA)
- Evolutionary Computation (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Physiology (AREA)
- Vascular Medicine (AREA)
- Databases & Information Systems (AREA)
- Psychiatry (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Fuzzy Systems (AREA)
- Geometry (AREA)
- Quality & Reliability (AREA)
- Pregnancy & Childbirth (AREA)
Abstract
Description
Claims
Priority Applications (6)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
KR1020227008925A KR102409090B1 (en) | 2017-05-11 | 2018-05-11 | Probability map-based ultrasound scanning |
CN201880030236.6A CN110753517A (en) | 2017-05-11 | 2018-05-11 | Ultrasound scanning based on probability mapping |
KR1020197035041A KR20200003400A (en) | 2017-05-11 | 2018-05-11 | Probability Map-based Ultrasonic Scanning |
EP18732986.7A EP3621525A1 (en) | 2017-05-11 | 2018-05-11 | Probability map-based ultrasound scanning |
JP2019561958A JP6902625B2 (en) | 2017-05-11 | 2018-05-11 | Ultrasonography based on probability map |
CA3062330A CA3062330A1 (en) | 2017-05-11 | 2018-05-11 | Probability map-based ultrasound scanning |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US201762504709P | 2017-05-11 | 2017-05-11 | |
US62/504,709 | 2017-05-11 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2018209193A1 true WO2018209193A1 (en) | 2018-11-15 |
Family
ID=62685100
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/US2018/032247 WO2018209193A1 (en) | 2017-05-11 | 2018-05-11 | Probability map-based ultrasound scanning |
Country Status (7)
Country | Link |
---|---|
US (1) | US20180330518A1 (en) |
EP (1) | EP3621525A1 (en) |
JP (1) | JP6902625B2 (en) |
KR (2) | KR20200003400A (en) |
CN (1) | CN110753517A (en) |
CA (1) | CA3062330A1 (en) |
WO (1) | WO2018209193A1 (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2020141908A (en) * | 2019-03-07 | 2020-09-10 | 株式会社日立製作所 | Medical imaging device, medical image processing device, medical image processing program and machine learning model |
US11986345B2 (en) | 2019-07-12 | 2024-05-21 | Verathon Inc. | Representation of a target during aiming of an ultrasound probe |
Families Citing this family (30)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109362221A (en) | 2016-03-09 | 2019-02-19 | 安科诺思公司 | Utilize the ultrasound image identifying system and method for smart network |
KR102139856B1 (en) * | 2017-06-23 | 2020-07-30 | 울산대학교 산학협력단 | Method for ultrasound image processing |
EP3420913B1 (en) * | 2017-06-26 | 2020-11-18 | Samsung Medison Co., Ltd. | Ultrasound imaging apparatus and control method thereof |
US11622744B2 (en) * | 2017-07-07 | 2023-04-11 | Massachusetts Institute Of Technology | System and method for automated ovarian follicular monitoring |
EP3777699A4 (en) * | 2018-03-30 | 2021-05-26 | FUJIFILM Corporation | Ultrasound diagnostic device and control method of ultrasound diagnostic device |
US11372086B2 (en) * | 2018-05-11 | 2022-06-28 | Qualcomm Incorporated | Radio frequency (RF) object detection using radar and machine learning |
US10878570B2 (en) * | 2018-07-17 | 2020-12-29 | International Business Machines Corporation | Knockout autoencoder for detecting anomalies in biomedical images |
WO2020016449A1 (en) * | 2018-07-20 | 2020-01-23 | Koninklijke Philips N.V. | Ultrasound imaging by deep learning and associated devices, systems, and methods |
EP3895600A4 (en) | 2018-12-11 | 2022-03-02 | Synergy A.I. Co. Ltd. | Method for measuring volume of organ by using artificial neural network, and apparatus therefor |
JP7192512B2 (en) * | 2019-01-11 | 2022-12-20 | 富士通株式会社 | Learning program, learning device and learning method |
WO2020150086A1 (en) * | 2019-01-17 | 2020-07-23 | Verathon Inc. | Systems and methods for quantitative abdominal aortic aneurysm analysis using 3d ultrasound imaging |
JP7273518B2 (en) * | 2019-01-17 | 2023-05-15 | キヤノンメディカルシステムズ株式会社 | Ultrasound diagnostic equipment and learning program |
US20200281570A1 (en) * | 2019-01-17 | 2020-09-10 | Canon Medical Systems Corporaion | Apparatus |
JP7258568B2 (en) * | 2019-01-18 | 2023-04-17 | キヤノンメディカルシステムズ株式会社 | ULTRASOUND DIAGNOSTIC DEVICE, IMAGE PROCESSING DEVICE, AND IMAGE PROCESSING PROGRAM |
JP7242409B2 (en) * | 2019-04-26 | 2023-03-20 | キヤノンメディカルシステムズ株式会社 | MEDICAL IMAGE PROCESSING DEVICE, ULTRASOUND DIAGNOSTIC DEVICE, AND LEARNED MODEL CREATION METHOD |
US20210045716A1 (en) * | 2019-08-13 | 2021-02-18 | GE Precision Healthcare LLC | Method and system for providing interaction with a visual artificial intelligence ultrasound image segmentation module |
CN110567558B (en) * | 2019-08-28 | 2021-08-10 | 华南理工大学 | Ultrasonic guided wave detection method based on deep convolution characteristics |
CN112568935B (en) * | 2019-09-29 | 2024-06-25 | 中慧医学成像有限公司 | Three-dimensional ultrasonic imaging method and system based on three-dimensional tracking camera |
US11583244B2 (en) * | 2019-10-04 | 2023-02-21 | GE Precision Healthcare LLC | System and methods for tracking anatomical features in ultrasound images |
US20210183521A1 (en) * | 2019-12-13 | 2021-06-17 | Korea Advanced Institute Of Science And Technology | Method and apparatus for quantitative imaging using ultrasound data |
JP7093093B2 (en) * | 2020-01-08 | 2022-06-29 | 有限会社フロントエンドテクノロジー | Ultrasonic urine volume measuring device, learning model generation method, learning model |
KR102246966B1 (en) * | 2020-01-29 | 2021-04-30 | 주식회사 아티큐 | Method for Recognizing Object Target of Body |
CN115916061A (en) * | 2020-04-07 | 2023-04-04 | 韦拉索恩股份有限公司 | Automatic prostate analysis system |
CN113616235B (en) * | 2020-05-07 | 2024-01-19 | 中移(成都)信息通信科技有限公司 | Ultrasonic detection method, device, system, equipment, storage medium and ultrasonic probe |
CN112184683A (en) * | 2020-10-09 | 2021-01-05 | 深圳度影医疗科技有限公司 | Ultrasonic image identification method, terminal equipment and storage medium |
KR102238280B1 (en) * | 2020-12-09 | 2021-04-08 | 박지현 | Underwater target detection system and method of thereof |
JP2023034400A (en) * | 2021-08-31 | 2023-03-13 | DeepEyeVision株式会社 | Information processing device, information processing method and program |
JP2023087273A (en) | 2021-12-13 | 2023-06-23 | 富士フイルム株式会社 | Ultrasonic diagnostic device and control method of ultrasonic diagnostic device |
JP2023143418A (en) * | 2022-03-25 | 2023-10-06 | 富士フイルム株式会社 | Ultrasonic diagnostic device and operation method thereof |
WO2024101255A1 (en) * | 2022-11-08 | 2024-05-16 | 富士フイルム株式会社 | Medical assistance device, ultrasonic endoscope, medical assistance method, and program |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20090264757A1 (en) | 2007-05-16 | 2009-10-22 | Fuxing Yang | System and method for bladder detection using harmonic imaging |
US20140052001A1 (en) * | 2012-05-31 | 2014-02-20 | Razvan Ioan Ionasec | Mitral Valve Detection for Transthoracic Echocardiography |
US20160350620A1 (en) * | 2015-05-27 | 2016-12-01 | Siemens Medical Solutions Usa, Inc. | Knowledge-based ultrasound image enhancement |
Family Cites Families (19)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5081933A (en) * | 1990-03-15 | 1992-01-21 | Utdc Inc. | Lcts chassis configuration with articulated chassis sections between vehicles |
JPH06233761A (en) * | 1993-02-09 | 1994-08-23 | Hitachi Medical Corp | Image diagnostic device for medical purpose |
US5734739A (en) * | 1994-05-31 | 1998-03-31 | University Of Washington | Method for determining the contour of an in vivo organ using multiple image frames of the organ |
US5984870A (en) * | 1997-07-25 | 1999-11-16 | Arch Development Corporation | Method and system for the automated analysis of lesions in ultrasound images |
US6238342B1 (en) * | 1998-05-26 | 2001-05-29 | Riverside Research Institute | Ultrasonic tissue-type classification and imaging methods and apparatus |
AU2001261163A1 (en) * | 2000-05-03 | 2001-11-12 | University Of Washington | Method for determining the contour of an in vivo organ using multiple image frames of the organ |
US20050089205A1 (en) * | 2003-10-23 | 2005-04-28 | Ajay Kapur | Systems and methods for viewing an abnormality in different kinds of images |
US8556814B2 (en) * | 2007-10-04 | 2013-10-15 | Siemens Medical Solutions Usa, Inc. | Automated fetal measurement from three-dimensional ultrasound data |
US8265390B2 (en) * | 2008-11-11 | 2012-09-11 | Siemens Medical Solutions Usa, Inc. | Probabilistic segmentation in computer-aided detection |
US20100158332A1 (en) * | 2008-12-22 | 2010-06-24 | Dan Rico | Method and system of automated detection of lesions in medical images |
US20110257527A1 (en) * | 2010-04-20 | 2011-10-20 | Suri Jasjit S | Ultrasound carotid media wall classification and imt measurement in curved vessels using recursive refinement and validation |
US9710730B2 (en) * | 2011-02-11 | 2017-07-18 | Microsoft Technology Licensing, Llc | Image registration |
US10292676B2 (en) * | 2011-12-21 | 2019-05-21 | Volcano Corporaton | Method for visualizing blood and blood-likelihood in vascualar images |
JP6323335B2 (en) * | 2012-11-15 | 2018-05-16 | コニカミノルタ株式会社 | Image processing apparatus, image processing method, and program |
JP6200249B2 (en) * | 2013-09-11 | 2017-09-20 | キヤノン株式会社 | Information processing apparatus and information processing method |
KR20150098119A (en) * | 2014-02-19 | 2015-08-27 | 삼성전자주식회사 | System and method for removing false positive lesion candidate in medical image |
US20180140282A1 (en) * | 2015-06-03 | 2018-05-24 | Hitachi, Ltd. | Ultrasonic diagnostic apparatus and image processing method |
DK3338642T3 (en) * | 2015-08-21 | 2020-07-06 | Fujifilm Corp | ULTRASOUND DIAGNOSIS DEVICE AND METHOD FOR CONTROLING THE ULTRASOUND DIAGNOSIS DEVICE |
US9536054B1 (en) * | 2016-01-07 | 2017-01-03 | ClearView Diagnostics Inc. | Method and means of CAD system personalization to provide a confidence level indicator for CAD system recommendations |
-
2018
- 2018-05-11 CA CA3062330A patent/CA3062330A1/en active Pending
- 2018-05-11 JP JP2019561958A patent/JP6902625B2/en active Active
- 2018-05-11 WO PCT/US2018/032247 patent/WO2018209193A1/en unknown
- 2018-05-11 CN CN201880030236.6A patent/CN110753517A/en active Pending
- 2018-05-11 KR KR1020197035041A patent/KR20200003400A/en active Application Filing
- 2018-05-11 US US15/977,091 patent/US20180330518A1/en active Pending
- 2018-05-11 KR KR1020227008925A patent/KR102409090B1/en active IP Right Review Request
- 2018-05-11 EP EP18732986.7A patent/EP3621525A1/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20090264757A1 (en) | 2007-05-16 | 2009-10-22 | Fuxing Yang | System and method for bladder detection using harmonic imaging |
US20140052001A1 (en) * | 2012-05-31 | 2014-02-20 | Razvan Ioan Ionasec | Mitral Valve Detection for Transthoracic Echocardiography |
US20160350620A1 (en) * | 2015-05-27 | 2016-12-01 | Siemens Medical Solutions Usa, Inc. | Knowledge-based ultrasound image enhancement |
Non-Patent Citations (2)
Title |
---|
IMANI FARHAD ET AL: "Ultrasound-Based Characterization of Prostate Cancer Using Joint Independent Component Analysis", IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, IEEE SERVICE CENTER, PISCATAWAY, NJ, USA, vol. 62, no. 7, 2 July 2015 (2015-07-02), pages 1796 - 1804, XP011584546, ISSN: 0018-9294, [retrieved on 20150616], DOI: 10.1109/TBME.2015.2404300 * |
P. LOONEY ET AL.: "Automatic 3D ultrasound segmentation of the first trimester placenta using deep learning", 2017 IEEE 14TH INTERNATIONAL SYMPOSIUM ON BIOMEDICAL IMAGING, 2017, pages 279 - 282, XP055758954, DOI: 10.1109/ISBI.2017.7950519 |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2020141908A (en) * | 2019-03-07 | 2020-09-10 | 株式会社日立製作所 | Medical imaging device, medical image processing device, medical image processing program and machine learning model |
JP7302988B2 (en) | 2019-03-07 | 2023-07-04 | 富士フイルムヘルスケア株式会社 | Medical imaging device, medical image processing device, and medical image processing program |
US11986345B2 (en) | 2019-07-12 | 2024-05-21 | Verathon Inc. | Representation of a target during aiming of an ultrasound probe |
Also Published As
Publication number | Publication date |
---|---|
US20180330518A1 (en) | 2018-11-15 |
CN110753517A (en) | 2020-02-04 |
KR20220040507A (en) | 2022-03-30 |
JP2020519369A (en) | 2020-07-02 |
EP3621525A1 (en) | 2020-03-18 |
KR20200003400A (en) | 2020-01-09 |
CA3062330A1 (en) | 2018-11-15 |
KR102409090B1 (en) | 2022-06-15 |
JP6902625B2 (en) | 2021-07-14 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US20180330518A1 (en) | Probability map-based ultrasound scanning | |
CN110325119B (en) | Ovarian follicle count and size determination | |
US11464490B2 (en) | Real-time feedback and semantic-rich guidance on quality ultrasound image acquisition | |
CN111629670B (en) | Echo window artifact classification and visual indicator for ultrasound systems | |
US10949976B2 (en) | Active contour model using two-dimensional gradient vector for organ boundary detection | |
US11432803B2 (en) | Method and system for generating a visualization plane from 3D ultrasound data | |
US11278259B2 (en) | Thrombus detection during scanning | |
US9324155B2 (en) | Systems and methods for determining parameters for image analysis | |
US10238368B2 (en) | Method and system for lesion detection in ultrasound images | |
Laporte et al. | Learning to estimate out-of-plane motion in ultrasound imagery of real tissue | |
CN112641464A (en) | Method and system for context-aware enabled ultrasound scanning | |
US20200196987A1 (en) | Method and system to manage beamforming parameters based on tissue density | |
US11684344B2 (en) | Systems and methods for quantitative abdominal aortic aneurysm analysis using 3D ultrasound imaging | |
CN112603361A (en) | System and method for tracking anatomical features in ultrasound images | |
KR20150103956A (en) | Apparatus and method for processing medical image, and computer-readable recoding medium | |
CN116258736A (en) | System and method for segmenting an image | |
US20190271771A1 (en) | Segmented common anatomical structure based navigation in ultrasound imaging | |
WO2020133236A1 (en) | Spinal imaging method and ultrasonic imaging system | |
EP3848892A1 (en) | Generating a plurality of image segmentation results for each node of an anatomical structure model to provide a segmentation confidence value for each node | |
WO2021230230A1 (en) | Ultrasonic diagnosis device, medical image processing device, and medical image processing method | |
JP2018157982A (en) | Ultrasonic diagnosis apparatus and program | |
JP2018157981A (en) | Ultrasonic diagnosis apparatus and program |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 18732986 Country of ref document: EP Kind code of ref document: A1 |
|
ENP | Entry into the national phase |
Ref document number: 3062330 Country of ref document: CA |
|
ENP | Entry into the national phase |
Ref document number: 2019561958 Country of ref document: JP Kind code of ref document: A |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
ENP | Entry into the national phase |
Ref document number: 20197035041 Country of ref document: KR Kind code of ref document: A |
|
ENP | Entry into the national phase |
Ref document number: 2018732986 Country of ref document: EP Effective date: 20191211 |